#78 - Leonard Kleinrock: Internet Creation, AI Future

29:42And then the researcher would say, but you see that graphics up in Utah? I want that here. And the high performance computer in Illinois, I want it here, et cetera, et cetera. Well, opposite. Now, wait a minute. We can't give everybody everything. But if you were in a network, you could log on to Utah and do the graphics there. Log on to Illinois and use the high performance computing there. So the concept in 1960 became, let's make a data network to connect together these really excellent, centers of excellence of computer science. So the idea of needing a network to connect computers was suddenly born in 1966 at ARPA.

30:23That was basically the government thread. Well, how to build it? I told you a moment ago, there was a theoretical thread which we've been cooking along, which we knew how to do it. So these two threads came together. And in 1966, they brought in one of my MIT classmates, Larry Roberts, to head up this effort at ARPA to create a data network, which is going to be called the ARPA Net, the ARPA Data Network. And Larry brought a bunch of us together from the theoretical side and from the government side to write a specification as to what this network should look like, send that specification out to industry, ask industry to bid on making a machine which would accomplish this technology, and deploy it across the country.

31:10So those two threads, in answer to your question, came together in 1966, the theoretical side, the government side, and now is a wonderful match. We understand how, we have a need, we got the funding, let's make it happen. So the plans began to lay out. And so how did you first conceptualize the idea of a global computer network? So the idea originally was just to connect together computers. And the original plan was to lay out a 19-node network across the United States, connecting together these industrial research centers and the university research centers. So the idea was, we're going to put these devices which implement the theory.

31:54You call those a router today. We call them a packet switch. and the particular name associated with the packet switch was called an interface message processor. These were standalone mini computers, which were going to be deployed across the country, connected together with high speed lines. But these computers would be running packet switches as opposed to circuit switches, which are being used for the telephone network, deployed these across the country and bring up the capability slowly. So the theory was there. We needed to implement it. So the specification went out to industry where we told industry what we needed to please bid on deploying a 19 node network.

32:41Well, it went out to industry, a number of companies bid, and a winner was selected around Christmas time of 1968, a company called Bolt, Baranek, and Newman, a research company, development company out of Cambridge, Massachusetts, made up of many researchers out of MIT Lincoln Laboratory, by the way, the same place that I got my degrees from. And they presented a terrific proposal. They added a lot of capability, a lot of functionality, a lot of protocols, and good engineering to this. And they were told that in eight months, by September, the Labor Days of 1969, you had to deliver the first one of these switches to UCLA, which would become the first node of the ARPANET.

33:28Now, why did they choose UCLA? Because in fact, we had the technology, the understanding, the theory, and the software and hardware people who could build this thing. So UCLA was supposed to become, and was, the Network Measurement Center to become alive in September, Labor Day of 1969. And that was the plan. And would you walk through your approach at that point? So the approach was BBN was going to deliver these things called imps, these routers, these packet switches. But we had to find a way to interface those switches to our host computers, which were going to be running on the network. So we needed to create something called a host IMP interface.

34:11And each host in a network on different sites was a different host. So each one needed their own interface. So at UCLA, we were busy creating the host IMP interface for our host, which has happened to be a scientific data systems Sigma-7 computer. Now that computer, I was running for the department here as a timeshared computer science research machine. And we developed the software, the Host IMP protocol. And I put together a hardware team, a software team, a research team of graduate students, and a staff, 40 people, many of whom were PhD students, to make this all happen. And by the way, the person I put in charge of the software group was a PhD student called Steve Crocker, who created what's called the Request for Comments series.

35:06he wrote the specification for the first Hostos protocol. And under him were people like Vin Cerf, well-known, John Postel, Charlie Klein, a number of other people, a really super group. So comes Labor Day weekend of 1969, the imp is delivered, we're going to turn it on the day after Labor Day, September 2nd, basically the Tuesday after the Monday Labor Day, we turn it on and we connect the imp and the host with a 15 foot cable. And we begin to move bits back and forth from the imp to the host. Now, who is there to watch this happen? Everybody. ARPA, BB &N, AT &T, we're going to use their long lines.

36:00GTE, we're going to use their local lines. Scientific data system, we're using their host. Honeywell, who was the manufacturer of the minicomputer that made the imp. UCLA, UCLA administration, ARPA, all the researchers. And everybody was ready to point their finger to the other guy if it didn't work. Well, happily, the bits began to move back and forth. But one node is not a network. We didn't have a network yet. And the plan was September UCLA, October Stanford Research Institute, 355 miles to the north would get their switch and connected their host. November UC Santa Barbara and December University of Utah, a four node network initially.

36:48So in October, SRI got their IMP and ARPA provided through BB &N the first high-speed piece of the backbone internet running from UCLA to SRI, 350 miles long, running at the blazing speed of 50 ,000 bits per second. Now, you wouldn't pay a nickel for that today, but in those days, that was high speed. So now we had the UCLA host connected to the UCLA switch, the IMP, connected to a high-speed line, connected 350 miles north to the IMP at SRI, connected to their host. We now had a two-node network. And now we're ready to test the functionality of what's supposed to go around. And what is the functionality?

37:41You sit at one computer, logged onto it. You log in through the network to a remote computer, log onto it, and use its services there. So we can run that test. So would you share your experience of the first message through the internet on October 29th, 1969? Exactly. So we're ready to make this test, the first two nodes. And we're going to send the first message over the ARPANET, which is going to become the internet. Now, what was that first message? Do we have a good one? Well, think about it. Samuel Morse, first telegraph message. He had a great message. It was, what hath God wrought? Biblical, prophetical.

38:23The first telephone message, come here, Watson, I need you. How about Neil Armstrong? Space, a giant leap from Antarctica. Those guys were smart. They understood the press, the public relations, the media. We were just a bunch of nerds. We were sitting there one night, ready to run this little test. I had one of my programmers, Charlie Klein, down here at UCLA. We had another programmer, Bill Duvall, up at SRI. And all we want to do is log in. We didn't have a message, we just want to log in. Now to log in, you have to type L-O-G. And that's going to go up to the SRI host. And the SRI host is smart enough to know, oh, they're trying to log in.

39:08It'll type the IN for you. Now understand, the SRI host has no idea that it's coming through a network. It thinks it's coming from a local user, a local timesharing user. Now when a local timesharing user connects to a timesharing computer, you type a key on the keyboard, it goes to the timesharing computer, which echoes that character back to you and prints it on your screen. And you keep repeating that. It's a very short distance, very fast. We're going 350 miles north. So we typed the L and just to make sure what's going on, you know, how do we know if the messages are getting up there, the characters?

39:48We set a telephone link between Charlie and Bill. So I set it up and Charlie typed the L and said to Bill, you get the L? And Bill said, yep. And it printed. Type the O, you get the O? Yep. Got the O. Type the G, you get the G. Crash. The system crashed. so a couple of comments what was the first message ever on the internet and the answer is low as in low and behold i added that later very wise but think of it a more prophetic more profound the more succinct message we couldn't have asked for we didn't plan it but it turned out to be beautiful now what crashed well there wasn't our host It wasn't our imp.

40:37It wasn't the high-speed line. It wasn't the SRI imp. It was the SRI host. Because the SRI host, I told you, when you type the G, it got smart enough to type the G-I-N back to you. And it wasn't set up to handle three characters. So there was a buffer overflow. It was a piece of patchwork that we put together, which quickly failed. We quickly patched it. And within the next hour, we got the full message too. So as you point out, October 29th, 1969, at 1030 at night Pacific time, the first message ever on the internet was low, as in lo and behold. And it went from UCLA to SRI, 350 miles north to the second computer on the network.

41:24And SRI is Stanford? It's Stanford Research Institute, which was not the university, it was associated with the university. What would you say was the original spirit and culture among the early internet pioneers? So that's a wonderful question. I'm going to harken back to Licklider. Licklider created that culture. When he went to the researchers, the Minsky's, et cetera, he said, look, we're going to give you this funding. We're going to let you do what you want. We're not going to bother you. Failure is okay. Shoot for the moon. You know, and then he gave that to his graduate students. What better research environment where you have trust in your colleagues, you are trusted, use ethics, this is free, it's open, it's shared, people are behaving well, and they're all trying to achieve a common, wonderful, difficult, challenging goal.

42:22It was really one of great camaraderie and working together. There was no sense of ownership, of patenting. Nobody patented anything. No intellectual property rights. Here it is. Our gratification was we create something, somebody uses it. That's about as good as it gets. And so it was a wonderful environment of cooperation. A research environment couldn't have been better. I told you what came out of that was graphics, networking, chip technology, artificial intelligence, on and on and on. from that wonderful open environment. But it didn't last that long because there's another element here which I haven't addressed and I didn't emphasize, but let me point out.

43:07ARPA was an agency created within the United States Department of Defense because that's where the funding was. But there was no defense, no military applications intended, at least not down at the level of the researchers. You know, we were doing a theoretical engineering job. But it wasn't too many years later when things like the Mansfield Act came in, which said, you can't fund this research unless there's somehow a military application somewhere down the line. And they also said, this is an unfair way to grant. We have to have a competitive environment. what's called a broad area announcement where multiple people get to bid for the project and for the funding.

43:56And to be honest with you, that sounds more fair, but to be honest with you, it slowed down the rate at which great work was done. It took a while to get the bidding. The contracts became smaller, more people bidding for yet smaller work. And it somehow lost that culture to which I referred and you asked about, which is a wonderful, open, shared, let's conquer the world kind of environment. It's similar to your early childhood where your mom let you do what you thought you wanted to do, and it was open. And then you fast forward to the early days of the internet, and you have the funding, and they say, shoot for the moon, do whatever you want.

44:42We're not going to oversee it. We just want you to create something great. and you go to smart people who are motivated and confident in their abilities and you give them the world to shoot for. And you start constricting it and you can see it leads down a very different path. It does. And there's a question of equity and fairness. Certainly you have to raise those issues, but that was the time for that open, free, flowing, flexible structure, because that was a remarkable golden era of creativity. I told you all the things that emerged and the people you work with, there was another aspect to it.

45:18Whereas the faculty gave the graduate students their free reign. Those graduate students, bless them, they formed their own network, their own network working group. Graduate students, not only at the university where they were getting their degree, but across the United States, across the world. They formed their own collaboration groups. They interacted, they grew in this easy, supportive, open, shared, let's solve a problem way. You have to understand what a wonderful, creative period that was. And it produced some remarkable results. Earlier, you talked about taking this idea, and I know you've written papers about building the internet way before it actually occurred.

46:05But you talked about this idea of taking it to well-established large companies like AT &T as an example. What type of resistance did you face and how did you persist through that? Great question, because you raise an actual broader issue. Along the way, were these ideas accepted by different constituents? And the first place where we met resistance was, as you say, I developed this mathematical theory for how to develop a data network. And so I, as well as Paul Barron, would go to AT &T and said, AT &T, build a data network. What an opportunity. We'll give you some idea how to do it. You have the wherewithal to do it, build it.

46:48And AT &T said, it won't work. And even if it does, we want nothing to do with it. Pretty haughty attitude. You know, AT &T didn't even bid on the contract that BBN won. Now, I thought about that. it was a big mistake. They really lost the foresight. But you think about it. Why were they so obstinate? And the answer is, in some sense, they were right. There was no data to send. There was no need at the time for a data network. There was no business model. So as a business, AT &T said no. Big mistake. But you can understand their thinking at the time. So the first resistance was that industry wanted nothing to do with it.

47:32We had to wait for the government to recognize they needed a network. And then here was it. I came to UCLA in 1963 with all this capability, but nothing to do with it. I kept doing research. Funny in 66, someone wanted to build it and off we go. Okay. So ARPA says, we're going to build a network. And what's it going to do? It's going to take various sites that they've been supporting and tell those sites, the MITs, the Utahs, the Illinois, Stanford, UCLA, join a network. Take your wonderful high-performance computers and put them in a network so other people share them. And the principal investigators of those sites say, what are you talking about?

48:16You're going to take my precious high-performance computer and steal some of my cycles for people out in the network? No way. So our own community resisted joining a network for understandable reasons. So Larry at the time and I went around to these sites and we said, look, don't you want to join a network? You'll be able to get access to other people's work. And so I went around and I said, look, if there was a network, how much of it would you use? One teletype, two teletypes worth? Yeah, one or two teletypes. And how much would you let the outside world use yours? Two teletypes worth. So I took those numbers.

48:57I went around to all the sites, 19 of them, and I created a traffic matrix. And I published it. And now these guys, in some sense, were committed. In June of 69, before the open that came up, I published that traffic matrix. But that still wasn't enough. What really clinched it was, ARPA went around to these sites and said, Mr. Principal Investigator, we are supporting your research. You will join the network. And of course they joined. So that was the second level of resistance. Our own community wanted nothing to do with it. So they joined early on. And that's a continuation of those two threads you described earlier.

49:42The theory side, and then the kind of the government support. and that kind of came together and that helped fuel the launch. Exactly. But when they get on now, how much use is there? These guys are reluctantly joining. Was it being used? The answer is no. And the answer is no, because in fact, it was very difficult to use this network. I mean, suppose I'm sitting at UCLA and I log on to Utah. When I need to log in, I need to know the command language. I need to know the applications and the services. That's a big learning process for me. So people were not willing to go through that learning process.

50:22In fact, one of the main things that went on is somebody would move, say, from Utah to UCLA and wanted to use the machine with which they were familiar, and they'd log on across the network. So it's very sporadic use. And meanwhile, we were testing the network, making it grow. However, as I mentioned before, there was something called a host-host protocol, a way in which hosts could communicate with each other easily. And Steve Krocker wrote the first RFC describing that. And we implemented that here at UCLA. The first host-host protocol, which was called the Network Control Program, NCP, was implemented basically in 1970.

51:05And so people began to use it a little bit. But it was decided that we ought to run a demonstration of these many sites that were coming up with their own great applications. So in October of 1972, we ran a demonstration of the ARPANET in the Hilton Hotel in Washington, D.C. Bob Kahn basically put it together. And he went around to all the sites and said, look, create a demo that you can run on this demo at the Hilton in October 72. There's going to be a major conference there. The conference was the International Conference on Computer Communications, 1972. And so we brought together a bunch of PIs at that demo.

51:57We rented out the basement of the hotel where this conference was taking place. We brought a package, which had imped there, ran some high-speed lines there, and people announced their applications. Many people brought chess playing programs. MIT brought a bunch of robots to run around the floor. Some people brought simulation, air traffic control, and set it up to run in October 72. And we invited not only the PIs, but this was a major conference. We invited the public in and showed them how to run some of the applications. Now, as a side, interesting story, what was the UCLA demo? The UCLA demo was that we're going to have John Pastel sit in Washington, log on to UCLA all the way across the country, pull up and execute a program here at UCLA, which required a photograph, which had reached across the country back to MIT to pull up.

53:03And then that picture had to be processed back in Utah. And when the processing done, send it back to Pastel in Washington and print it at the local printer. That is a pretty good demo. Log on to UCLA, back to MIT, across to Utah, and back to Washington. So John was practicing it that night. He logged on, sent the picture from Utah to his printer, but nothing printed. And he looked around the demo room and it turns out that the robots that MIT had brought were jumping around the floor. The printer output had gone to the robots. We fixed that real fast. So you can see these demos were quite exciting.

53:48And it grew an enormous amount of interest, not only on the part of the public, But we, principal investigators, saw what's capable now. We saw what was out there. So that launched a significant increase in the use over the net. And that was a turning point in the open-ed use. But again, the reluctance was there all the way until we demonstrate ease of use, functionality, and something useful. So would you talk through how the internet transitioned from a research tool to a commercial and social platform eventually. That's a scary story. And one that's in everybody's face today. So let's talk about it.

54:30Okay. 1969, we bring up the network. We get a four-node network by the end of 69. By middle of 70, we have a 10-node network or spanning the country. And the network continued to grow and beautifully, quickly, easily across the country with nice applications. And what was the driving force? The engineering aspect to it. Adding functionality, capability, making it faster, better, easier to use. Pushing the boundaries of technology. It was a research engineering project. And that was fine until 1988. In 1988, the first worm, the first virus got released. The first broad-based virus got released by Robert Morris, a graduate student at Cornell.

55:27He released it and he claims his office mate did it. This virus went out and it basically infected a large number of computers across the country. And we looked at that and we said, ouch, what's going on? Then we said, uh-oh, it's just a hacker messing around. What a mistake. That action was a harbinger of the dark side of the network, which was about to emerge. But we ignored it. And what's interesting is, Robert Morris' father was an employee at the CIA at the time. And he announced at the time, it's a good thing my son released this, because it's a warning. And we said, what are you talking about?

56:14He was right. So from 1988 to 1994, the network continued to grow and some very important things happened. Number one, Al Gore. Al Gore recognized a report that I had shared for the National Research Council, talking about something called a National Research Network, was of interest to him. He had me testify before one of his Senate subcommittees, which I described what a nationwide research network would look like. He convinced the first George Bush to create what he then called the information superhighway, the gigabit backbone. And that was thanks to Gore making it the High Performance Computing and Communication Act of 1991.

57:03So the first event that occurred in this period between 88 and 94 was, we now have a backbone, a gigabit backbone with capability. Second thing, in early 1980s, NSF began to deploy supercomputers around the country. And by the tail end of 1988, 1980s, they wanted to connect these supercomputers together. and what better network than the ARPANET to connect them together. But the ARPANET wasn't fast enough. So NSF increased the speed first to one and a half megabits per second and then to 45 megabits per second and created something called the NSFNET. Now, when NSF came into the picture, the constituency of the players in the ARPANET changed.

57:55Instead of just computer scientists, we now had scientists. Chemists, physicists, biologists, oceanographers, psychologists, a much broader community. Now, where does a research chemist work? As an example, either at a university or a chemical research laboratory in a large chemical company. So picture that. You got this research laboratory in a large chemical company on the ARPANET. And these guys are doing what? What are those chemists doing? They're using email. This email was the most seductive application at the time. Now, this email is being used inside this boundary of the research group in the company.

58:40But the rest of the chemical companies, oh, that email looks interesting. The staff, the management, the owners, they see this and they want it. So suddenly, in the late 80s and early 90s, a demand for dot-coms began to emerge. Now we have a demand, we have a capability, we've got a backbone network which can support this, but what's missing? What's missing is a simple user interface. Well, in the early 90s, what happens? The World Wide Web appears. A simple to use graphical user interface. And suddenly, all this combination comes in and now it began to reach out to the consumer world, to the general world.

59:27And now, so suddenly consumers are on it, companies are on it. And around this time on April 12th of 1994, another critical event occurred. The first broad-based spam message was launched. It reached most of the people on the network. It was launched by two lawyers on April 12th, 1994. and what it was, was a message going out. And by the way, I've got a copy of that message, that email message. And it said, was reaching out and said, look, there's a green card lottery coming up. We will help you get in the lottery. Come to us, hire us, pay us. We'll let you get in. Those lawyers were advertising on our research network, on our engineering.

1:00:20That's not allowed. We were aghast. We said, ouch. And this time we said, uh-oh. So we sent email back to those lawyers. We said, you can't do this. How dare you? Shame on you. Cease and desist. We sent so much email back to their server that we took down their server. So an unintended consequence of the first broad-based spam message was the first denial of service attack. But it was too late. It was too late. the commercial world saw that here was a way to reach the consumer public, they realized this is not a research engineering network. This is a shopping mall. This is a social network. This is an entertainment channel.

1:01:05What wonderful capability to reach out and make this into a commercial success. So in answer to your question, it was around that time that the focus and the energy going into the development of the now called the internet shifted from research and engineering to commercialization. The network took a significant shift to the left. And now the energy was how to seduce the consumer to spend their money. And we've seen that development continuously now. And of course, that drove the directions and the energy in the wrong direction. At the same time, since we enabled so many people to come on, we brought in the power of the internet.

1:01:53What is the power of the internet? The power is that anybody with a computer and aligned to the internet, no matter how poor or dirty or banana peels on the floor in a dirty environment, can reach out at almost no cost instantly to millions of people and influence them, connect with them, et cetera. Now that's the power of the network and it's also a perfect formula for the dark side of the network. And so it began to emerge. We began to see these terrible things come onto the internet, you know, fraud, denial of service, fake information, et cetera. And around that time, as it began to emerge, I said, oh my goodness, the internet is going to its juvenile teenage years and it'll mature.

1:02:49Didn't happen. It didn't happen. The internet is now in some sense in a worse situation because the social networks began to dominate and became influencing what was going on in the information, the misinformation, the motives, the directions. And it's risen in some very nasty ways. And now it's even more serious. Now it's not only nuisance hackers and a nuisance. We've got nation states who are putting boundaries around their internet. And when they do that, you lose the free access across the world. We've got organized crime. We've got people who are committing fraud, serious fraud across the network in organized ways.

1:03:38And the social networks and the bubbles that are being created on the internet, as you know, are devastating. And it's very hard to control these things. And we don't have the mechanism in the internet at this point to try to prevent that. So an answer to your story is we've taken a ride from the time that spam message came out in a direction which is really a really unsettling situation right now with the internet. What would you say are the most surprising or significant changes as the internet has evolved? We've been able to predict very effectively the infrastructure of the internet, high-speed networks, basically capability in the walls, smart spaces, wireless networks, devices, and high-speed networks and connectivity, which works.

1:04:36What we've not been able to predict well are the applications and the services. We didn't predict email coming. It came on and suddenly within months dominated the traffic of the internet. We didn't see peer-to-peer networks. We didn't see user-generated content like YouTube. We didn't see blockchain coming in. We didn't see search engines. We didn't see shopping malls. We didn't see all of these things that have come about and the social networks. To go back to social networks, I'm going to take us back a bit and I'm going to answer a question you didn't ask. And that is, when was the concept, the vision of what we now have as the internet first articulated?

1:05:21When did someone see this? And I can ask people listening to this to wonder, is there anyone they can think of or any person? Well, I'm going to quote somebody and I'm going to ask you to think about when and who this could have been. The quote said, essentially, it will be possible for a businessman in New York to reach out across the ocean using a device no larger than a watch instantly at almost no cost to his colleague in London or elsewhere and send easily any picture, drawing, text, speech immediately. Now, first of all, whoever that was, they're talking about what we now call the internet.

1:06:14So who do you think that was? And when do you think it was said? Well, I know the answer because I've studied it. So I don't want to spoil the surprise. Well, it is a surprise. It was Nikola Tesla. And he said that in 1908, more than a century ago. And he was talking about the telegraph network. He didn't talk about video because there was no video, but he had the concept. Along the way, people like H.G. Wells, Vannevar Bush, Lick Leiter, even myself articulated visions, much of which has come true, some of which hasn't. But my point is, the vision was there, but it had to wait for the technology to catch up before it could be implemented.

1:06:57And that happened around 1969, when communications technology was sufficient enough and chip technology was fast enough to allow these rapid switching to take place. And as a little anecdote, months before the imp arrived at UCLA, UCLA put out a press release and that press release was an interview with me, which I'm quoted as saying word to the effect that these computer networks, once they're deployed, will be always on, always available, everywhere and anybody with any device can get on at any time. And it will be as simple to use as electricity and it will be as invisible as electricity. So in some sense, I was predicting web-based IP services.

1:07:51I was predicting simple access. I was predicting great technology, lots of use. But the one thing that I totally missed totally missed, was social networks. I wasn't talking about computers talking to each other, or people talking to computers, but not people to people. That was something we totally missed. And one thing I predicted would just still not happen is the invisibility. Electricity is a fantastic service. It's two plugs in the wall. You plug in, you get electricity. You don't need a password. You don't need a sign-on. You don't need all those complications. plug in, you get it. The internet is not that easy.

1:08:35You need passwords, you need keyboards, you need small things, you need permissions, you need wifi, et cetera. Well, it's slowly becoming invisible as we have deployed technology in our walls, integrated circuits, et cetera. But it's happening. But my point is there were a number of visions way back when, and now we finally articulated it. What's fascinating about the history that you just shared. It took about 60 years since Tesla made that description until the internet was effectively born. And here we are almost 60 years later from that point, and it's still not invisible. So parts of the technology move fast and parts of it move incredibly slow.

1:09:15Yes. And you raise another interesting point. The internet, before it took this bad turn, had more than 20 years, 25 years to be curated, to be refined, to be improved, to be structured. Some of these newer technologies come out, are coming out and hitting us very quickly without a proper curation time. I'll give you one example, blockchain. Blockchain just came out and it came out with a dollar sign in its mouth, which basically has corrupted the idea of a distributed ledger and it's really broad applications. And there are other technology which are coming out very quickly without proper curation time.

1:10:02So if you had the power to rewrite history and you could go back and change one thing about how the internet developed, what would it be? I would recognize the potential for abuse of the internet. And I would have put in early on some proper protection, some control. And two things I surely would have done, and was partially done but not followed up, was first of all, to have strong user authentication, to have some way to prove that the person that's communicating with you is who he or she claims to be. Some way to prove that. Right now, the man in the middle kind of thing is a serious problem.

1:10:42you could be talking to someone and could be impersonating some third party. So strong user authentication. And the second thing would have been strong file authentication, that the file I just delivered to you is the one I sent you and has not been corrupted. Some way to prove that. Now, we had some simple ways to do that early on, but we never followed it up in some serious ways. Why not? Because the community we were dealing with was trustworthy. We were honest. We had a common goal in mind. We weren't sending false information or impersonating people. So had we installed those early protections, the first thing we should have done was turn them off.

1:11:25And then as the need arose, crank them up slowly. But that still would not have protected us against all of the abuses we have today. The really bad actors, it's hard to control them. And I'm not sure what we could have done back then to prevent evil, foul-minded people from abusing what we have today because criminals have always been with us since time immemorial. And it's the few bad actors that can cause the broad spread and very effective impact on the good systems of the world. So you just highlighted some of the downsides of the internet, but obviously it's had profound impact over the last several decades.

1:12:09Would you say overall you're satisfied with where the internet stands today or not? Now, satisfied is a powerful word. I'm happy with the way it's evolved because it's given some enormous good. I mean, anybody can gain access to the world's education. We're giving people for free 4 ,000 years of knowledge, easily accessible, of great value to them at no cost. And we've allowed people to interact in ways they never could before. can reach out across the world from your basement and reach out to the great minds and great colleagues and great companions. On the other hand, the dark side is there polluting some of this stuff.

1:12:51So I'm really unhappy about that, but I'm thrilled at the wonders that the internet has provided. And it's changed society in many ways for the better, in some ways for the worse. And we are facing a serious challenge now is how to correct and adjust some of the serious problems we basically opened up. Are there any lessons in particular that you feel future innovators should draw from the Internet's history? Yes, to look forward to the potential dangers of the technologies we're unleashing. And lots of technology are being unleashed right now. Look, there's quantum, there's fusion, there's blockchain, there's AI, and all these sort of new wonderful technologies coming out.

1:13:42And I'm not sure how much forethought that is into where those could lead down bad pads that need to be corrected and anticipated now. because many of these new technologies are not going through the 25 years of curation. They're exploding out there. People are saying, oh, let's launch it. Let's use it. Oh, how wonderful, but not how terrible it can become. What do you see as the future of the internet and where do we go from here? The future of the internet is interesting because new magnificent technologies are now joining into it. The most prominent is AI. AI is emerging very quickly. It started out back in the 1950s, late 50s, with the Marvin Minsky's and the John McCarthy's doing symbolic AI.

1:14:32And now it's all about large language models and neural networks of great capability with great concerns, because you don't know how they work. That's a whole other story. But I think that AI is going through its early, very rapid growth stages. and people are concerned about the ethics and the dangers and the unleashing of what could be uncontrollable functionality. So people are thinking about it, but I also fear that it's being launched far more quickly by powerful forces of industry that are not as concerned with the dangers, but more with the how to exploit it. Now, in my altruistic academic mind, I like to think that AI itself may be the solution to some of the problems that AI itself is creating, that some of the dangers it's creating, that some of the dangers that the internet itself is created, like to watch over, to curate what's going on there in the way the AI systems are being deployed and what they're given access to.

1:15:48AI is very powerful and it may be powerful enough to control itself. Now, that's a wild dream. I don't have the solution, but I think it has the potential to be the answer to the problems that we as humans are having great difficulty with, be it on the internet, be it AI, be it some of these other applications. And that's just a pipe dream, but it's an optimistic feel. I wanted to ask you a few questions about AI. But first, if we kind of close a chapter on the internet, looking back, do you feel that the discovery and development of the internet stands among the most transformational inventions in human history?

1:16:28I think it does. I think it was inevitable. Yeah. It certainly has changed every aspect of humanity, be it social, technology, entertainment, education, just industry, commerce. It's enabling great strides and great capability, great efficiencies and great dangers. And the fact when you get something so powerful is exactly when you have to worry about the way it can be abused and misused. and therefore we need the great minds of today to add some sanity to the way in which these things are being deployed. And to be honest with you, the tech industry is not the best example of people who are monitoring in the right way.

1:17:16They have a profit motive. They're there to make money. They're there to make their shareholders happy. And I worry, I fear and worry that the driving forces are not there to curate in the proper way. So let me ask you a few questions about the parallels and lessons from the internet as it relates to AI. How do you think the current wave of AI innovation compares to the internet's early days in terms of vision, risk-taking, and also collaboration? So in terms of genealogy, both have had a very long history. But the way AI has evolved, it's gone from a particular approach called symbolic AI, where we were understanding how things work and building functionality that we could observe, control, and predict.

1:18:14once we got to the neural networks, networks which are graded doing facial recognition, image recognition, and many other things, we don't understand how they perform. We're trying to provide some explainability for these networks, but it's not yet here. So look at the situation. We have these neural networks, which are extremely capable and they're optimized to do certain things, which brings me to the related point. Let's talk about optimized systems, of which neural networks are an example. Whenever you have an optimized system, you've got systems that perform very well in the domain for which they were designed.

1:18:59And I'm going to give you an example, something we all understand, AM radio versus FM radio. AM radio is terrible all the time. And as you move from the base station, the further you move, the lousier it gets. FM is terrific. And it starts out with wonderful reception. One, until you reach the boundary for which it was designed, and then it collapses badly. Now, if you know where the boundary is and you avoid it, that's good. But if you have a system which is performing well and doing some important things, and you don't know what the boundary of its functionality is, you're in serious danger.

1:19:39and I'll give you a particular example from the past in AI. Many years ago, there was a program written, an AI program written to play checkers. A computer will play checkers with you. And it was using rote learning, and the developer of that program would play games with it. And one day, the developer changed the algebraic sign of the objector function from plus to minus by accident. The machine was now programmed to lose, not to win. And when he started playing with it, the scary part is he couldn't tell it was trying to lose. Because in order to lose well, you've got to get control of the board and make the big sacrifice.

1:20:23And he didn't know it. Okay, now let's race forward to today. Let's take a neural network program whose job it is to increase and improve the economy of the United States. And it's optimized to do that. And it's happily doing it. If it reaches the boundaries of which you could do it well, it's going to collapse. Or if it suddenly decides or someone makes a mistake and says, now you want to ruin the economy in the United States, you won't know it until perhaps it collapses. So optimized systems, which we don't understand how they work, and we may not know what the boundaries of their limits are, are extremely dangerous.

1:21:06And in my sense, that's what neural networks are today. They're very capable. They have a range over which and a domain over which they work well. And we don't know how they work. And we're trying to fix that and trying to understand them. And by the way, they're great for many things. You know, they can detect the way the ocean waves are fluttering and explain what's going on underneath the ocean surface in ways that humans never could. But there's a fear there. And until we get to understand these things a little better, I worry about deploying them willy-nilly in many domains of importance.

1:21:43One of the most fascinating aspects of what you described in terms of the early days of the internet is the environment that was created where you had governments come in with funding and giving you the free reign to shoot for the moon and be innovative. Do you feel like AI research today, the environment for it, fosters the same kind of bold foundational work that enabled the internet to thrive? It's enabling open-ended research and serious and powerful funding. Guidance, I'm not sure how much guidance there is, but the thing that worked for the internet was no control. Let them find air views of great interest and value.

1:22:29So whether the AI research community will continue to find valuable, powerful, and beneficial functionality is an open question. But the capabilities, there's a lot of money going into it, a lot of independent research is going in different directions. So it's got the wherewithal and similar in some sense to what the internet did. But the guidance was not there for the internet and I don't think it's there for AI. Are there lessons from the internet's commercialization phase that are most relevant to AI's current trajectory? Yes. The lesson that we learned badly in the internet was we didn't anticipate the dark side.

1:23:11Now I think we're smart enough to anticipate the dark side of AI and try to protect against it. So the similarity is there, and I think we have the advantage of knowing where we're heading. But it's an uncertain road. I mean, this is a frontier. And do we have only benevolent players? No. We have, again, rogue organizations and countries that have mischief in mind. And it is a danger. I mean, it's not unlike nuclear proliferation. How do you control that kind of a thing? You know, and the consequences of not controlling are severe. And we've been very bad at both. How can the AI community avoid repeating some of the similar mistakes, such as short-term thinking or loss of public trust?

1:24:05Well, I think we need to get the various stakeholders to engage in proper discussions. And in terms of the evolving internet, I can think of at least four stakeholders. The first set of stakeholders are the scientists who create the technology. Second set of stakeholders are the commercial folks who implement it and deploy it. The third stakeholder is the government. And the fourth set of stakeholders are you and me, the user community. Now, the scientists are trying not only to develop new technology, but trying to find protective mechanisms against the dark side. It's up to us to try to find homomorphic encryption, secrecy, et cetera, which will protect us against some bad things.

1:24:57And we're moving in that direction. But the technology world, the commercial world, they've got to cooperate and they've got to provide the ability for people to express their concerns about what's going on. Now, I'll jump ahead. What's the role of the government? The role of the government mainly is to provide a forum whereby the stakeholders can discuss these issues and maybe provide some oversight, some large level oversight that industry cannot do on its own. But the group that's doing the least is the user community. The user community is not complaining or speaking up about the abuses to which they're subject.

1:25:41You know, your privacy is being invaded. Your assets are being stolen. You're being asked to sign legal documents you don't understand. And only lately are we beginning to see industry say, here's the privacy policy I'm going to apply to you. But they describe it in a way you can't understand. there should be a way to describe the privacy policy that you're being subject to in some graphical way. This is what it looks like. You got so much of this, a little bit of this, less of this. And you should have a picture of what you were willing to do. And if what they're offering doesn't match what you want, you say bye-bye or you say modify or negotiate.

1:26:23That kind of interaction among the stakeholders has to take place. And I don't see yet a rich environment whereby that discussion is taking place among all these various stakeholders. So in terms of that kind of advice taking place, not only to the internet, but for the people moving into these newer technologies like AI, like blockchain, like fusion, like quantum, needs to take place across the board. We have to bring in all the players and have all of them express what they're thinking of, what they're afraid of, what solutions they can offer. Now that's pie in the sky, but I think that's the mindset we need going forward instead of individual thrusts in individual directions for individual gains.

1:27:09That's going to lead us down a very dangerous path. You've described your vision for the internet as invisible to users like electricity. An example that you brought up, do you see AI becoming as seamlessly embedded in daily life? Yes, I do. and it's part of the internet being invisible. I like to imagine, as I did in that vision I had back in July of 1969, I should be able to walk into a room and the room should know I've walked into it. And I should be able to interact with the room the way you and I are talking now, namely with speech, with gestures, with facial expressions, with haptics, with the way you and I interact without having a keyboard or technology or one of these damn things, tiny keyboards.

1:28:03And it should know what my privileges, my profiles, my preferences are. So it can enable, can provide them to me. If I walked up to a physical device, it should be enabled with those capabilities that I want. And it should anticipate what I want, be able to be an agent for me, make suggestions, and interact the way humans do, using all of the internet capability, but the AI tech capability as well. I want an agent with whom I can interact, which is very much like interacting with you. And maybe it is you with capability, enhancing you in your interaction with me. You know, I foresee, I like to say, what we're moving to is a global interface, intelligent interface surrounding us.

1:29:04It's just a, but it's in the ether. It's in the environment. It's in our tables, our cars, our walls, our fingertips, our bodies. and the environment in which we engage. And sometimes technology can advance faster than society can accept that advancement. So you kind of need both to line up. You do, and typically they don't. Typically they don't. To me, it feels like technology, that the pace of advancement is so fast. I just generally feel like it's probably ahead of where society on average is comfortable. Do you agree with that? I do. And you can create a great new widget and deploy it and then figure out what it can do and be surprised what it can do without anticipating those uses.

1:29:56And some of those uses are not what you want. But you're right, technology is moving very quickly. But that's been the nature of human civilization from time immemorial. You discover iron, things change. You discover the wheel, things change. You discover the internet, things change. Are there unique challenges or opportunities that AI presents compared to the internet? Yes. As I said, one of the scary things about AI is we don't know how the latest versions work. And will they escape to a domain that we can't even control? Will they start running things well beyond our ability to control them?

1:30:39The internet typically did not do that. It was people interacting with the internet, which made its functionality and its applications. AI can go beyond that, I think. The idea of robots is also in the picture there. How do they behave? What kind of morality and ethics do you instill in them? Who decides what they are? It's a little worrisome as to how we're going to manage to... I hate to use the word control. but help direct the way those things move into the future. It's a challenging time. For sure. Where do you see the greatest opportunities for AI to complement or even extend the Internet's legacy?

1:31:29The ability to understand information and events, to discover new mathematics, to discover new principles that maybe we as humans have been having trouble with or are unable to, to put together ideas. Bringing together information to create new recognition is one of the marvels of human civilization. AI has ability to do that quicker than we can, more effective than we can, but hopefully in ways that are benevolent. But the ability for a neural network or a large language model to have all the knowledge that's out there and understandable and accessible and composable is something that we as humans don't have.

1:32:26And by the way, that brings me to another point, which so much goes backwards in time, is the following. I think that computers are the worst enemy of critical thinking. And I say that because we relegate too much to the computer. We don't put into our head things that we should know. For example, how many people really know what Archimedes' principle is or Maxwell's equations or understand physics or understand how things work? and they just relegate it, even do arithmetic. They relegate it to a computer. And if it's not up here, that means you can't think with it. You can ask a computer to think with it, but when you take a shower or drive a car or fall asleep, you want ideas percolating in your head to generate new ideas.

1:33:20And if it's not there, if you relegate it to the computer, you've lost. So the computer is great for remembering and handling things that it does well, but you can't take it all out of your head. And I find that's the case very often with some of my graduate students right now. I'll give them a mathematical model to work out. And they'll come back a few weeks later and show me Y versus X. They've simulated the performance of some system I've asked them to. And I'll look at the performance. I'll say, oh, that looks interesting. That looks like a straight line. What does the slope of that line mean in terms of your model?

1:34:01No idea. Why is the asymptote that value? Don't know in terms of the physical model. And then I'll ask him another question. I say, what if I double some parameter? What will that curve look like? And their answer is, I'll simulate it again and figure it out. I say, no, no, no. You need a model to be able to explain to me what it will be. and once they come up with the result, say they do get a great result, they don't ask questions like, what is that result telling me? Is there a principle that's trying to expose to me? Can I use it somewhere else? It's the why that I asked before that the physicist asked.

1:34:44What is there to learn out of it? They don't ask that question, they just move on. And so they're down a level of understanding. Largely because they've got the computer around to address some of these issues. You can rely on that crutch and overly rely on it. Yes. Yes. And then we don't make great minds. That's right. And so is your sense that we're entering a new golden age of innovation today? Yes and no. If the innovation comes out of these computers and these AI systems, where's the human? If the human goes along and collaborates in this process, then we have a winning situation, I think.

1:35:26But if you relegate and just let it run on its own, you're giving up the human element and you're giving up a lot of what could be great, great innovation and move into a world where you don't know what the heck's going on. And that's a scary scenario. I think of it as humans are good at certain things. Computers are good at certain things. And the two together could be very powerful. But if you as a human just rely on the computer for all the things that it does, and you give up the things that you're good at, you can easily be replaced by computer. You're exactly right. In fact, go back to Licklider, the man-computer symbiosis.

1:36:09Engelbart said the same thing. He had this augmentation system. put humans and computer systems together and they'll augment each other. And that augmentation system is extremely powerful. And we don't want to give that up. And there's a danger that we do if we just relegate it all to the artificial world. Well said. Well, Len, you've spent a lot of time with us. You've shared about a century of internet vision and history, development, evolution, and provided some glimpse into the future with AI and future technologies. I appreciate all the insights you shared with us. It was fascinating for me and I hope for our listeners as well.

1:36:50My pleasure. Thank you. Thanks for listening. We hope you enjoyed this episode. Please visit our website at insightfulinvestor.org to access past shows and learn more about our podcast. If you have questions, feel free to email us at info at insightfulinvestor.org. And if you enjoyed the discussion, please subscribe to this podcast to ensure you don't miss future episodes. And don't forget to forward today's conversation to others you think would enjoy listening. This podcast is provided for informational purposes only and should not be relied upon as legal, business, investment, or tax advice.

1:37:28All opinions expressed by podcast participants are solely their own opinions and do not necessarily reflect the opinions of Evoke Advisors, their affiliates, or companies featured. Due to industry regulations, participants on this podcast are instructed not to make specific trade recommendations, nor reference past or potential profits. And listeners are reminded that securities trading, commodity trading, and alternative investments are complex and carry a risk of substantial losses. As such, they are not suitable for all investors.

1:38:01Listeners should be aware that guests featured on The Insightful Investor may have current or past associations with Evoke Advisors or the host, including as an investment manager of a private fund opportunity by Evoke or access through an affiliated Evoke fund or as a client. Participation as a guest on the podcast should not be perceived as an endorsement or testimonial with respect to Evoke Advisors, the podcast host, or their services. Similarly, the inclusion of a guest on the podcast does not imply that Evoke Advisors or the host endorses the guest or any company with which they may be affiliated or employed.

1:38:40Evoke has neither paid nor received compensation from guests for their participation.